Who was at fault, and why — decided in seconds
TAM × Najm reconstructs collisions from photos, dashcam footage, a driver's voice note and written statements — then apportions liability with regulation-grounded reasoning an adjuster can defend.
See
Damage geometry, skid marks, signals, lane position
Hear
Voice statements transcribed and cross-examined
Judge
Fault split with cited traffic articles
—
Live ledger
—
Awaiting first case
< 60 sec
vs 11 days manual
—
Across filed cases
Case throughput
Collision mix
Modality lift
Mean engine confidence by the evidence present in the retrieval context.
Case ledger
LiveNo cases filed yet. Every verdict you run is stored here with a shareable dossier.
Operations control
Live worklists with confidence routing — approve, return for clarification or send a file to specialist investigation without leaving the dashboard.
Recommendations waiting on a Najm approver
Nothing in this worklist.
Evidence intake
Fuse scene imagery, dashcam video, a spoken statement and written notes into one liability verdict.
Evidence is fused across vision, speech and language before apportioning fault.
No active reconstruction
Add at least one piece of evidence above. The engine performs best with a wide scene shot, a close-up of the damage, and the driver's own account.
What we do
A liability engine that reads the crash the way a forensic investigator would
Most claims stall because fault is argued from partial evidence — one photo, one account, one adjuster's reading. TAM × Najm fuses everything captured at the roadside into a single reasoning pass, and returns an apportionment with the reasoning attached.
Scene & damage vision
Crush geometry, impact angle, resting positions, debris fields, skid marks, lane markings and signal state are read from photos and plate-level close-ups.
Dashcam video reasoning
Footage is sampled into key frames across the pre-impact window so the engine sees the manoeuvre, not just the aftermath.
Voice statement analysis
The driver or witness speaks in their own words. The engine transcribes, weighs and cross-examines the account against the physical evidence.
Written account parsing
Adjuster notes, police summaries and claim text are fused into the same retrieval context as the imagery and audio.
Regulation-grounded verdicts
Every apportionment cites the MOI Traffic Law article or Najm schedule row it rests on, so the decision is defensible on appeal.
Adversarial stress-test
One click builds the opposing driver's strongest rebuttal and tells you whether it would actually move the fault split.
Inside the analysis
The reasoning an adjuster can defend in front of a claimant
Damage geometry tells you the direction of travel
Crush depth, paint transfer and the height of the impact line separate a genuine rear-end from a reversing collision, and a sideswipe from a lane intrusion — before anyone's statement is weighed.
Statements are cross-examined, not believed
Every claim in the driver's account is checked against what the photographs physically support. Contradictions are listed with a severity, so an adjuster sees exactly where the story breaks.
Fraud signals are separated from missing paperwork
Pre-existing rust in a 'fresh' impact, staged debris, damage inconsistent with the described speed — those raise the fraud score. An untaken statement or a missing exhibit is a completeness gap, never an accusation.
Fault rules, applied categorically
Where Saudi traffic law is categorical — following too closely, failure to yield on a left turn, reversing out of a parking bay — the split is 100/0 with the article cited, not hedged into a soft 70/30.
The sequence is reconstructed second by second
The engine produces a timed event list and an overhead sketch of the road layout, vehicle headings and point of impact, so the reasoning can be checked at a glance.
Confidence you can act on
The score names the physical findings driving it. High confidence means damage geometry, statement and fault rule all agree; a low score tells you which single element is uncorroborated.
How a case moves
Ingest
Photos, dashcam clips, a voice note and written notes land in one intake. Video is auto-sampled; audio is captured in-browser.
Fuse
Multimodal RAG puts every modality into a single retrieval context alongside Saudi traffic regulation, so signals corroborate or contradict each other.
Reconstruct
The engine rebuilds the second-by-second sequence, names the manoeuvre error and explains why the mistake was made.
Adjudicate
Fault is apportioned to 100%, scored for confidence, priced for settlement and filed as a shareable dossier you can interrogate.